What Is Google Antigravity 2.0? Agent-First Dev Platform (May 2026)
What Is Google Antigravity 2.0? Agent-First Dev Platform (May 2026)
Antigravity 2.0 is Google’s bet that agents — not IDEs — are how software gets built next. Announced at Google I/O 2026 on May 19, 2026, it’s a complete agent-development ecosystem: desktop app, CLI, SDK, Managed Agents in the Gemini API, and enterprise integration. The default model is Gemini 3.5 Flash, the launchday Flash-tier frontier-class model.
Last verified: May 20, 2026
Quick facts
| Property | Value |
|---|---|
| Announced | May 19, 2026 (Google I/O 2026) |
| Vendor | |
| Default model | Gemini 3.5 Flash |
| Platforms | Windows, macOS, Linux (desktop); CLI; SDK; API |
| Pricing | Free download, pay-as-you-go for API calls |
| Previous version | Antigravity 1.0 (late 2025) |
| Replaces | Workflows that previously used Cursor-style “code with AI” |
| Powered by | Gemini 3.5 Flash + Gemini 3.1 Pro (selectable) |
The five pieces of Antigravity 2.0
| Component | What it is |
|---|---|
| Antigravity 2.0 (desktop) | Standalone app on Windows/macOS/Linux. Central hub for agent interaction, multi-agent orchestration. |
| Antigravity CLI | Terminal-native surface for spawning agents. For developers who don’t want a GUI. |
| Antigravity SDK | Programmatic access to the same agent harness behind Google’s internal products. Host on your own infra. |
| Managed Agents (Gemini API) | Pay-per-call Antigravity agents via the Interactions API. No infra to manage. |
| Gemini Enterprise Agent Platform | Google Cloud-native deployment for enterprise. Connects directly to GCP projects. |
What’s new vs Antigravity 1.0
- Manager view — orchestrate multiple agents across multiple workspaces from one pane. Editor view (the IDE-like surface) is still there but no longer the default.
- Parallel multi-agent execution — spawn N agents on different tasks; review when they’re done.
- Dynamic subagents — agents can spawn child agents for sub-problems with their own context windows.
- Scheduled / background tasks — agents that run on cron or on triggers.
- Native voice commands — speak to the manager view.
- Antigravity CLI — new in 2.0.
- Antigravity SDK — new in 2.0. Lets third parties embed the agent harness.
- Managed Agents in Gemini API — new in 2.0. The Interactions API endpoint.
- Gemini 3.5 Flash as default — replaces 3.1 Pro as default; faster, cheaper agentic loops.
- Firebase + AI Studio + Android Studio — tighter integration into the broader Google dev stack.
Antigravity 2.0 vs Cursor 3 vs Windsurf
| Antigravity 2.0 | Cursor 3 | Windsurf | |
|---|---|---|---|
| Vendor | Anysphere | Codeium | |
| Architecture | Agent-first (manager view) | IDE-first with agent mode | IDE-first with Cascade agent |
| Default model | Gemini 3.5 Flash | User picks (Opus 4.7 popular) | Codeium model + frontier picks |
| Multi-agent | Native, parallel, manager view | Multi-agent panel (newer) | Cascade (single agent + tools) |
| CLI | Yes (Antigravity CLI) | Limited | Limited |
| SDK / embeddable | Yes (Antigravity SDK) | No | No |
| Background tasks | Yes (scheduled) | Limited | Limited |
| Enterprise platform | Gemini Enterprise Agent Platform | Cursor for Teams | Windsurf for Teams |
| Open ecosystem | MCP + Google tools | MCP | MCP |
| Pricing | Free + API usage | $20/mo Pro, $40/mo Business | $15/mo Pro |
Honest read: Antigravity 2.0 is the most opinionated agent-first product among them. Cursor 3 and Windsurf are catching up on multi-agent, but their default is still “you’re driving the IDE.” Antigravity’s default is “agents are driving; you’re reviewing.”
Who Antigravity 2.0 is for
- Engineering teams doing high-leverage refactors, migration work, test coverage backfills — the kinds of tasks where you want N agents in parallel.
- Indie devs who want to ship features overnight with a fleet of agents.
- Platform / DX teams at companies building on Google Cloud — use Managed Agents in the Gemini API directly.
- Anyone running on Vertex AI — Antigravity is the agent layer on top of Vertex’s model catalog.
Pricing
The desktop app is free to download. The actual cost is model usage through the Gemini API. Practical guidance:
| Plan | Antigravity capacity |
|---|---|
| Free Gemini | Low limits — fine for trying it |
| Google AI Pro ($19.99/mo) | Standard limits |
| Google AI Ultra ($99.99/mo) | 5x Pro limits + priority |
| Google AI Ultra top ($199.99/mo) | 20x Pro limits |
| Vertex AI (pay-per-token) | No subscription, billed per call |
Limits and caveats
- It’s a big behavior shift — going from “I write code, AI helps” to “agents write code, I review” is real change-management work for teams.
- Multi-agent debugging is still hard. The manager view helps but parallel agents can produce overlapping diffs.
- Default model is Flash — that’s good for speed and cost but if you want maximum quality you’ll want to switch to Gemini 3.1 Pro (or 3.5 Pro when it ships in June).
- Outside Google’s stack — you can use Antigravity with non-Google models via the SDK, but the default tooling assumes the Gemini API.
TL;DR
Antigravity 2.0 is Google’s bet that 2026 is the year the IDE gets demoted from primary to backseat. Multi-agent, manager-view-first, free to download, billed per API call, and aligned around Gemini 3.5 Flash for fast cheap agentic loops. If you’re a Cursor or Windsurf user evaluating where 2026’s dev stack is going, this is the one to spin up this week.